Search

• Jointly consider the task dependency and edge cloud environment. • Combine deep reinforcement learning and genetic algorithm. • Generate the initial population of genetic algorithm by DQN. Edge computing can greatly decrease the delay between users and cloud servers, which can significantly improve system service performance. However, it remains challenging for more efficient scheduling and allocation of users' application demands with dependence constraints to edge cloud servers. Due to the randomness of the initial population, traditional intelligent optimization algorithms have poor convergence speed in addressing resource scheduling. Therefore, to minimize the execution time of the application, this paper proposes a hybrid algorithm to solve the resource scheduling problem with parallelism and subtask dependency. To improve the convergence speed of the algorithm, this paper makes full use of the features of deep Q networks (DQN) and genetic algorithms (GA). The initial population of GA is generated using DQN. Finally, to evaluate the effectiveness of our proposed algorithm, this paper selects three real scientific workflows for experiments. The experimental results show that the hybrid algorithm can converge quickly and improve the optimization effect in a short time. [ABSTRACT FROM AUTHOR]

• Supervised feature selection of high-dimensional data is formulated as an MOP. • We developed a problem-specific non-dominated sorting genetic algorithm to solve the MOP. • We made a systematical comparison between our method and some state-of-the-art FS approaches. Feature selection (FS), which plays an important role in classification tasks, has been recently studied as a multi-objective optimization problem (MOP). In this paper, we consider minimizing three objectives of FS and propose a problem-specific non-dominated sorting genetic algorithm (PS-NSGA). In PS-NSGA, an accuracy-preferred domination operator is applied, which makes the individual with higher classification accuracy in the population more likely to survive. And a quick bit mutation is used, which breaks through the limitation of traditional bit string mutation and increases the efficiency. In addition, a mutation-retry operator and a combination operator are designed to make our algorithm converge faster and better. At last, a solution selection strategy is developed to determine the most proper feature subset from the obtained Pareto solutions. Experimental results on 15 real-world high-dimensional datasets demonstrate that our proposed algorithm can achieve competitive classification accuracy while obtaining a smaller size of feature subset compared with some state-of-the-art evolutionary and traditional FS algorithms. [ABSTRACT FROM AUTHOR]

Association rule mining is a popular data mining task, which has important in many domains. Because the task of association rule mining is very time consuming, evolutionary and swarm based algorithms have been designed to find approximate solutions. However, these approaches still have long execution times, especially when applied on dense and big databases, or when low minsup and minconf threshold values are used. Moreover, these approaches suffer from the lack of diversity in the rules presented to the user. To address these drawbacks of previous algorithms, this paper proposes an efficient parallel algorithm named CGPUGA. It is a genetic algorithm that runs on clusters of GPUs to efficiently discover diversified association rules. It benefits from cluster computing to generate rules. Then, to evaluate rules, which is the most time consuming task, the designed algorithm relies on the massively parallel GPU threads. Furthermore, to deal with the issue of rule quality, the search space of rules is partitioned into several regions assigned to different workers, and rules found by each workers are the merged to ensure diversification. The designed approach has been empirically compared with state-of-the-art algorithms using small, medium, large and big datasets. Results reveal that CGPUGA is 600 times faster than the sequential version of the algorithm for big datasets. Moreover, it outperforms state-of-the-art high performance computing based association rule mining algorithms for real big datasets such as Pokec, Webdocs and Wikilinks. In terms of rule quality, results show that the designed CGPUGA algorithm provides rules of higher quality compared to the state-of-the-art NIGGAR, MSP-MPSO and MPGA algorithms for diversified association rule mining. [ABSTRACT FROM AUTHOR]

The paper shows how to use a genetic algorithm to design quasi one-dimensional structures with given properties. The superlattices were surrounded by water and made of epoxy resin and glass, with a layer thicknesses selected in such a way that a phononic bandgap occurs in the frequency range of acoustic waves. Multilayer transmission was calculated using the Transfer Matrix Method algorithm. In order to determine the optimal objective function for the genetic algorithm, the entire space of the possibility of the layers' distribution for a six-layer structure was analyzed. The spread of the transmission integral values even for the 6-layer structure was significant and ranged from 11.9% to 81.5%. Minimizing the value of the transmission integral as an objective function could lead to a large number of high transmission peaks with a small half width. It allowed the determining of the optimal objective functions for minimizing transmission in a given frequency range, and another allowing for its maximization. The phase diagram of the solution space for the transmission integral from the integral of the absolute value of the transmission functions derivative was determined. Transmission of the most and the least optimal six-layer structures was shown depending on the assumed objective function. Then, the analysis of fifteen and thirty-layer structures was carried out. From the dependence of the objective function values for the best individuals for each generation, it was determined that despite carrying out a thousand iterations, the stable state was determined before reaching 250 iterations of the algorithm, and thus the optimal structure was determined. The transmissions ratio of analyzed structures in the studied frequency range (up to 20 kHz) to full transmission (without structures) in this acoustic frequency range for the analyzed optimal 6, 15 and 30-layer structures were 16.78%, 9.6% and 4.41%, respectively. • The use of a genetic algorithm to optimize the structure was investigated. • It is necessary to select the objective function for the expected transmission. • The transmission depends on the amount of superlattice layers. • Increasing the number of layers reduces transmission. [ABSTRACT FROM AUTHOR]